The Agent Skills Directory

[PROMPT_INJECTION]: Indirect prompt injection surface exists in scripts/sequential_think.py. The skill breaks a question into steps and then feeds the output (reasoning and conclusions) of one model call directly into the prompt of the next call.
Ingestion points: Untrusted data enters the context through the chat() function which returns model-generated text from OpenRouter.
Boundary markers: Absent. The content from prior steps is concatenated into the context_str variable and inserted into the prompt string without delimiters or instructions to ignore embedded commands.
Capability inventory: The script performs network operations (requests.post) to the OpenRouter API.
Sanitization: None. The extract_conclusion function only performs basic string splitting and does not filter for malicious instructions.
[COMMAND_EXECUTION]: The SKILL.md file suggests running Python scripts using {baseDir} placeholders, which is a standard execution pattern for this environment.

sequential-thinking